v1.6.0

Changed

• The default chunk_size for AsyncGzipBinaryFile and AsyncGzipTextFile is now 256 KiB (was 64 KiB). This improves bulk-read throughput (~1.3–1.6x) and lets CPU-bound zlib work offload to a thread by default, at the cost of more buffer memory per open file. Pass chunk_size=64*1024 to restore the previous footprint.
• CI now runs on Windows and macOS in addition to Linux, and enforces a coverage floor. Fixed the mypy configuration for newer mypy releases.

Performance

• Bulk read(-1) no longer copies decompressed output through an intermediate buffer, speeding up full-file reads (notably for compressible data).
• Text read(size) and long-line readline()/iteration are now O(n) instead of O(n^2) for large reads.

Documentation

• Documented that an open file is not safe for concurrent use by multiple tasks (use one file object per task).
• Refreshed the performance guide for the new 256 KiB default and current benchmark numbers.